Embed Size (px)
Citation preview
ASTROMETRIC MEASUREMENTS OF SATELLITES WITH
DYNAMIC VISION SENSOR
Michał Żołnowski(1)
, Krzysztof Kamiński(2)
, Tobias Delbruck(3)
, Alicja Gąsiorowska(2)
, Justyna
Gołębiewska(2)
, Marcin Gędek(1)
, Monika Kamińska(2)
, Joachim Kruger(2)(4)
, Mikołaj
Krużyński(2)
, Mikołaj Pieniowski(1)
, Rafał Reszelewski(1)
, Krzysztof Romanowski(4)
, Edwin
Wnuk (2)
, Wojciech Dimitrow(2)
, Wojciech Mich (4)
(1)6ROADS Sp z o.o., Godebskiego Street 55a, Kraków, Poland, [email protected] (2)Astronomical Observatory Institute, Faculty of Physics, Adam Mickiewicz University, Poznań, Poland,
[email protected] (3) Sensors Group, Institute of Neuroinformatics, UZH-ETH Zurich, Winterthurerstr. 190, CH-8057 Zurich,
[email protected] (4)ITTI Sp z o.o., Rubież 46, Poznań, Poland
ABSTRACT
The Dynamic Vision Sensor (DVS) technology, which
could potentially be utilized in various SST activities, has
been the subject of our research over the last several
months. This type of camera records from each pixel an
asynchronous stream of events of brightness variation
changes, rather than integrated intensity recorded
simultaneously over selected exposure time. Due to its
novel approach to visual data retrieving and processing,
DVS needs evaluation as a potential tool in the SST
domain. Several observing methods have been proposed
and tested. The key parameters considered in our studies
are the camera’s timing accuracy and astrometric precision. The paper presents the results of the world’s
first astrometric measurements of satellites observed with
a DVS camera performed on a GNSS satellite.
Furthermore, a series of observations of principally GEO
and MEO, but also LEO, targets was conducted.
Satellites were observed with a few different optical
telescopes to evaluate DVS camera’s performance in
different configurations. A whole observing
methodology and reduction pipeline has been developed,
including dedicated GNSS clock for event timestamping
and dedicated software capable of converting DVS data
stream from native AEDAT format to a series of FITS
images.
1 INTRODUCTION
Dynamic Vision Sensor technology has been suggested
as an interesting alternative to CCD and CMOS cameras
several years ago (see [1] or [4]). Its primary advantage has been pointed out as a huge dynamic range allowing
for daytime detections. Also the principle of operation of
DVA cameras seems to fit perfectly for satellite survey
observations, with its low latency and sparse low data
rate. On the other hand the DVS technology is relatively
new, prototypical, uncooled and lacks large high-
sensitivity sensors with fully employed capabilities for
low light applications. In this project we decided to use a
first back-side illumination hybrid DVS and imaging camera prototype, namely DAVIS 346BSI developed by
University of Zurich.
2 DVS SPECIFICATIONS
The DVS camera, which underwent thorough
examination was DAVIS 346BSI. The pixel size of this
camera is 18.5μm x 18.5μm, the array size of 6.4mm x
4.8mm, 100% fill factor and the peak quantum efficiency
of 93%s. Three copies of the camera were assembled for
the tests. Fig.1 presents the DAVIS 346BSI camera.
Figure 1. DAVIS 346BSI camera.
3 HARDWARE & SOFTWARE
PREPARATION
By default, the camera’s firmware timestamps each
recorded event with microsecond resolution using its
internal oscillator without reference to any external
source. The beginning of each recording (event stream)
is time-stamped with PC control software (jaerViewer)
Proc. 8th European Conference on Space Debris (virtual), Darmstadt, Germany, 20–23 April 2021, published by the ESA Space Debris Office
Ed. T. Flohrer, S. Lemmens & F. Schmitz, (http://conference.sdo.esoc.esa.int, May 2021)
which can be referenced to UTC via NTP with a typical
accuracy at the level of 10 milliseconds. LEO tracking
requires accuracy at the level of sub-milliseconds and
millisecond accuracy is required for higher orbits. In
order to meet those accuracy requirements and to provide
the time reference independent from Internet connection
delays, external time impulses are injected into the DAVIS 346BSI camera in periodic manner. A dedicated
GNSS based clock has been constructed to generate these
impulses with microsecond accuracy. They are recorded
by DAVIS 346BSI camera as so called “special” events,
which are time-stamped in the same way as normal
events, however they can be easily distinguished. Two
time measurements of the same impulse, one by GNSS
clock, one by DAVIS 346BSI camera timestamp enable
the conversion of the camera’s internal clock to the UTC
time scale. Fig.2 presents the scheme of DAVIS 346BSI
camera events timing system.
Figure 2. DAVIS 346BSI camera events timing system.
Subsequently, it was necessary to develop software,
which would enable the effective use of the data provided
by the DAVIS 346BSI cameras with a regular
astronomical reduction and analysis pipelines. The software – AEDAT Converter – performs two tasks
simultaneously. First – it transforms event data stream
from AEDAT format to FITS images by counting all
events in each pixel between selected start and stop time
and saving them as integer ADUs. User is able to select
single or multiple time segments and event types to be
saved. The second functionality is to linearly interpolate
time passage between recorded “special” events and
convert internal DVS timestamps into UTC for each
event individually. Overall this software allows to select
into how many images each recording is converted and
therefore it controls the timing resolution of resulting
image series up to 1 microsecond, which is the limit of
DVS timestamps precision. Fig. 3 presents the scheme of
the process of converting timestamps to UTC.
Figure 3. Conversion of timestamps to UTC
4 LABORATORY ANALYSIS
A series of laboratory test with artificial, moving light
sources were conducted prior to observations in order to
determine a set of optimum DVS camera settings. Two
teams using different approaches found a slightly
different combination of settings that were later used
during observations. The results are presented in Tab.1
Parameter Default
settings
6Roads
Night
6Roads
Daylight
AO
AMU
Night
AO
AMU
Daylight
DiffBn 5.1n 115.5n 31.7n 51.7n 51.7n
OnBn 194.5n 317.6n 95.7n 244.6n 486.3n
OffBn 124.6p 0 0 97.9p 49.3p
PrBp 7.9p 6.4p 6.0p 7.9p 7.9p
PrSFBp 8.0p 4.6p 4.3p 8.0p 8.0p
RefrBp 593.6p 5.6n 5.1n 593.6p 593.6p
Table 1. DAVIS 346BSI camera settings used during the
night-time and daytime observations.
In order to better understand DVS camera behaviour in
low light observations a simple test was conducted to
measure the level of delays of two simultaneous light-
source blinks. Using our DAVIS 346BSI camera
equipped with a GNSS clock we recorded blinks of a LED, controlled with another GPS receiver in order to
compare the actual time of LED light up with the
recorded time. Additionally, by observing also a LED
reflection from a piece of glass we had the opportunity to
compare reaction time at about 25x lower illumination,
as illustrated in Fig. 4.
Figure 4. Testing setup for timing of a blinking LED
experiment.
Figure 5. DAVIS 346BSI camera recording of a blinking
LED and its reflection at three time intervals. In the top
image no events are recorded even though a new second
has started. In the middle image first events from
blighter light source is detected, but no events from its
reflection. In the bottom image the dimmer light source
blink is finally detected.
DVS setup Bright source
(direct LED)
Dim source
(reflected LED)
Nighttime 0.0029s ±
0.0006s 0.015s ± 0.003s
Updated nighttime
On Bias 1.32x
higher
0.0046s ±
0.0014s 0.020s ± 0.005s
Daytime
On Bias 2.27x
lower
0.0034s ±
0.0006s 0.017s ± 0.004s
Table 2. DAVIS 346BSI delays in recording blinking
LED and its 25x dimmer reflection.
As presented in Fig. 5 and Tab. 2 we detected measurable
delays in all recordings, but the 25x dimmer light-source
had a significantly larger delay. This result shows that
DVS camera time bias is mostly negligible in case of
GEO satellites observations, but significant in low orbit
targets. It also shows that DVS camera can have different
time bias for simultaneously observed objects if their
brightness differs significantly, which can be difficult to
correct for during analysis.
5 DVS OBSERVING METHODOLOGY
In both satellite survey and tracking observations, the
DAVIS 346BSIcamera struggles to detect stationary
objects with 0.3m telescope — unless their magnitude is
brighter than about 8. New observing techniques were
developed in order to overcome this limitation.
5.1 Rotating shutter
The fundamental rule of the observations performed with
the use of an event based sensor is the need for changes
in the field of view. Tracked objects, such as stars or
satellites do trigger events only because of scintillation
or tracking errors, thus they have low SNR when
compared to objects that move.
The initial solution devised by the project’s members was
employing a regular astronomical focuser in order to
make stellar images out of focus and then rapidly focus
them. Rapid refocusing would create an effect similar to reappearance of all objects in the field of view. However,
this method proved to be inefficient.
Further tests consisted of quick covering and uncovering
the end of the scope, which was aimed towards the sky.
The dramatic increase in flux from zero, when the scope
was covered, to full aperture when it was opened,
triggered a clearly visible cascade of star-caused events.
Eventually, a commercial, 5-positional colour filter
wheel produced by ASI ZWO was modified to
periodically cover and evenly uncover the DAVIS’ chip.
Subsequently, it was installed between the camera and
the telescope.
Figure 6. The modified CFW prior to attachment to the
camera with modified blades of the shutter.
The filter wheel was modified to spin continuously at a
user-selectable rate using an Arduino board. During
observations, the filter acts as a rotating shutter, similarly
to the shutters used with meteor cameras. Although the
initial results of using this method were promising, it was
noted that better darkening of the shutter’s interior was
necessary as stray light worsened the overall performance
significantly. Therefore, entire interior of the wheel’s
chamber as well as the connector were covered with
black material, resulting in improved results.
In further stages of the evaluation of the rotating shutter
method it was revealed that it improves limiting
magnitude for stationary objects to a large extent,
however, not exactly to the same level as telescope
movement. The test was performed when the Moon was
full, thus the bright background could have had a
negative impact on the results, which is worth noting.
Figure 7. Observations of Alcyone star region in
Pleiades with shutter method. The events triggered just
after the blade of the shutter-like wheel has uncovered
the DAVIS surface. The brightness of stars: (a) 2.85m,
(b) 8.52 m, (c) 6.29 m, (d) 8.97 m, (e) 8.71 m.
Figure 8. Observations of Alcyone star region in
Pleiades during slewing of the mount with angular
velocity of 240 arcsec/sec. Visible waves of ON and
afterward OFF events generated by the movement
toward the lower-right corner. The brightness of stars:
(a) 2.85m, (b) 8.52 m, (c) 6.29 m, (d) 8.97 m, (e) 8.71 m.
5.2 Epicycle tracking
Typically, satellite tracking is based directly on its
ephemeris. When the ephemeris is sufficiently accurate
and the telescope equatorial mount is accurate enough,
the target satellite is kept constantly at the same camera
pixels. In tracking with DVS camera this significantly
reduced (by about 3-4 mag) the limiting magnitude and
thus a different approach has been developed.
The satellite ephemeris was augmented with a circular
motion, with the period of 6 seconds and the radius
ranging from 6 to 15 arcseconds. As a result, the telescope was constantly wobbling around the ephemeris
position of the satellite and satellite image was moving in
the field of view with constant velocity. By manipulating
the radius and period it was possible to determine at what
angular speed the satellite was going to be moving during
tracking. It was found that the best results can be obtained
when the radius is 6 arcsec and the angular speed of the
satellite is about 6 arcsec/sec.
The drawback to this method is the fact that it introduces
complex, wave-like trails for reference stars, instead of
simple, linearly elongated images. However, this issue can be solved by using a relatively long rotation period
dividing the data into time steps small enough for the
curvature of the stellar trails to become undetectable.
Moreover, the AO AMU’s astrometric software is
capable of processing targets of any shape.
Figure 9. An example of “epicycle tracking” with a
radius of 15 arcsec. This image shows “on” events
registered while tracking GPS satellite no 45854. The
telescope is tracking in such a way that the satellite is
making small circles (slightly deformed because of
cos(Dec) and telescope interia). A single star passing
through the field of view is also visible.
Figure 10. A comparison of the same satellite
observations made at the same time with DAVIS 346BSI
camera using different tracking modes. All images
represent On events recorded during 1 sec. In the top
image a satellite survey observation with a telescope in
sidereal tracking mode is presented. A GPS satellite
traveling at about 45 arcsec/sec is clearly visible. In the middle image a satellite tracking observation is
presented. The satellite is only barely detectable and
only in some images (one of the best is presented here).
In the bottom image an “epicycle tracking” mode is
presented. The satellite is clearly visible, at the level
comparable with survey mode. Also a field star is
recorded.
6 OBSERVATIONAL TESTS
6.1 GPS satellite tracking
In order to comprehensively compare DVS and sCMOS
camera performance in satellite tracking a dedicated
observation campaign was conducted between January
and March 2021. The observations were conducted using
two identical 12.5 inch f/8 telescopes. One of the
telescopes was equipped with a DAVIS 346BSI camera.
The other telescope was equipped with Andor Zyla 5.5 camera. Both cameras were connected with dedicated
GNSS clocks for accurate image timing.
The software used for astrometry required at least 4
reference stars for a reliable solution. However, due to
the narrow field of view of the DAVIS 346BSI camera as
well as the satellite magnitudes rarely high enough for
detection even when moving in the field of view, it was
nearly impossible to meet this requirement using regular
satellite tracking. As a result, the two novel techniques
developed by 6ROADS and AO AMU, described in
Section 5, were employed, as well as numerous small
incremental improvements in reduction and analysis
software.
On the 3rd of March 2021, the most successful
observations were conducted. The observed target was a
GPS satellite (NORAD 40294, COSPAR 2014-068A).
The telescope equipped with the DAVIS 346BSI camera
collected events from 22:33:45 to 22:54:57 UTC. The
data stream was converted to 2472 FITS images, each
containing the “on” events solely and covering the
timespan of 0.5s.
Figure 11. An example of GPS satellite tracking
observation with DAVIS 346BSI camera and CDK12.5
f/8 telescope. The image is composed of “on” events
recorded during 0.5 seconds. The satellite is not clearly
visible here.
Figure 12. An example of GPS (40294) satellite tracking
observation with DAVIS 346BSI camera and CDK12.5
f/8 telescope. The image is composed of “on” events recorded during 6.0 seconds. This is presented only to
show where the satellite was observed (inside the red
circle). The telescope epicycle tracking with the satellite
constantly making small (R=6 arcsec) circles is visible.
The converted data was then analysed with Poznań
Satellite Software Tools. Astrometric data were
automatically extracted from 8 FITS files obtained from
DAVIS 346BSI and from 992 FITS files obtained from
Andor Zyla. The difference reflects the significantly
smaller field of view of the DAVIS 346BSI camera and
approximately 2 mag worse limiting magnitude. Rarely
more than 3 stars with SNR higher than 2 were observed simultaneously for the software to identify the observed
field. Fig.14 presents the comparison of selected
astrometric positions derived using DAVIS 346BSI and
Zyla cameras.
Figure 13. Comparison of selected astrometric position
of GPS (40294) satellite derived on 08-03-2021 using
DAVIS 346BSI and Zyla cameras.
All astrometric positions derived with both cameras were
compared with high accuracy SP3 ephemeris provided by
the GNSS operator. In order to derive the both regular
and random shifts in RA and Dec, the differences were
analysed statistically. Moreover, the difference along and
across satellite orbits was calculated in order to have a
better understanding of the nature of observed
inaccuracies. The systematic differences along orbit were
associated primarily with image or event timing
systematic errors and the random differences across orbit
were associated with overall astrometric accuracy. Large
systematic differences across the satellite's orbit should
not be observed unless there is an error in data reduction or in ephemeris. Small differences might be present due
to, for example, noncentral target placement in images or
systematic asymmetries in stellar images due to imperfect
collimation, filed rotation etc.
Figure 14. Distribution of errors in RA and Dec for
GPS (40294) satellite observed with DAVIS 346BSI
camera.
Figure 15. Distribution of errors along and across the
trajectory of a GPS (40294) satellite observed with
DAVIS 346BSI camera.
As seen in Fig.15, the data obtained using DAVIS
contains small but systematic deviations from ephemeris
positions. In the direction across satellite trajectory (Fig. 16) it can be noticed that the astrometric uncertainty is at
the level of 0.6 arcsec, which can be considered a very
good result. Moreover, no significant systematic
deviation across orbit can be noticed. All of the
systematic errors are oriented along the trajectory of the
satellite, which indicates that they are most likely
connected with event timing errors. Almost all of the
measured points are concentrated around -125 ms time
bias. With the satellite's angular velocity of about 40
arcsec/sec this corresponds to a shift of 5 arcsec or about
3 pixels in DAVIS camera. Several potential primary
explanations of this shift have been considered.
According to the first one, the shift could be attributed to
the slow reaction time of the DAVIS camera for very dim
targets, such as the observed satellite. This effect was
measured before in laboratory experiments and expected
to produce a time shift at the level of about 20ms for
target that were brighter than the satellite. Since the
camera reaction time drops with target brightness it can
be suspected that in case of nearly undetectable light
source the delay will be significantly larger. This effect
affects also all reference stars and creates flux dependent
distortions in relative X,Y positions of stars and satellites.
Another factor, which could potentially have an impact
on the shift is the asymmetry of the images of reference
stars. Each of the recorded moving object contains a
clearly visible “comet tail”, produced by pixels, which
trigger “on” events with significantly longer than usual delays. It seems that pixel triggering has the effect of
increasing the probability of producing random events of
the same polarity immediately afterwards. The
asymmetry was partially mitigated by including higher
weights to pixels with more events but we suspect that
some residuals might still influence the results. It is hard
to estimate the order of magnitude of this effect, however
if we assume that it can shift image by 1 to 2 pixels than
it is slightly too low to explain the -125ms time bias. The
final hypothesis was that the DAVIS 346BSI camera has
an intrinsic time bias, which is considered probable as
most cameras used in satellite tracking exhibit time bias
at the level of tens or hundreds of milliseconds. It was noticed multiple times that the camera control software
displayed a warning saying that “a non monotonic even
has been detected” continuously throughout the
observations. This could indicate that there were certain
timing issues related to the camera’s firmware and
hardware.
Similar analysis was performed for the images of the
same satellite obtained simultaneously using the Andor
Zyla camera on an identical telescope. Its result shows a
time bias of -18 ms and astrometric uncertainty at the
level of 0.15 arcsec (see Fig. 17 and 18).
Figure 16. Distribution of errors in RA and Dec for
GPS satellite observed with Andor Zyla 5.5 camera.
Figure 17. Distribution of errors in RA and Dec for
GPS satellite observed with Andor Zyla 5.5 camera.
Figure. 18 Top-left: Image composed of “on” events and cross-section line of star trail recorded with the DAVIS
346BSI camera. Top-right: distribution of “on” events along the cross section line. A trail of “on” events is visible to
the right. Bottom: Image and cross section of star trail recorded with the Zyla camera
Overall the results for DAVIS 346BSI camera have
acceptable and repeatable time bias and astrometric
accuracy on par with our expectations given that Andor
Zyla had significantly better limiting magnitude and
more reference stars.
In order to further confirm the results of the observations,
the orbital elements of the observed GPS satellite were
determined using data from both cameras. The software
used for this process was the GEODYN 2 package. The
number of astrometric measurements with Andor Zyla
was about 100 times larger than with DAVIS 346BSI
camera. As all data points in GEODYN 2 have the same
weights, the influence of DAVIS data points on the
resulting orbit is very low. Due to this fact, the results
should be interpreted only as a relative comparison of
DAVIS 346BSI measurements with respect to Zyla.
Figure 19. Orbital fit residuals in RA.
Figure 20. Orbital fit residuals in Dec
In Fig. 20 and 21, the blue dots representing the
measurements from DAVIS 346BSI are deviated by a
few arcsec because of its larger time bias. We can also
see larger scatter because of its lower astrometric
uncertainty resulting primarily from lower number of
reference stars.
6.2 Ajisai satellite
Ajisai is an experimental satellite on low Earth orbit
designed specifically for optical monitoring with ground-
based laser stations and telescopes (NORAD - 16908,
COSPAR - 1986-061A). It is a rotating sphere composed
mostly of mirrors and retroreflectors. The satellite was
clearly visible to the DVS during the specular reflections
of sunlight.
The Ajisai satellite was observed with DAVIS 346BSI
camera on a 12.5 inch f/8 telescope. With this setup, the
satellite was observed 2 times in total — on the 20th of
November 2020 for 120 seconds and the 13th of February 2021 for 125 seconds. The third observation of the Ajisai
satellite was performed using a 12” Newton telescope
and lasted 44 seconds. The event streams from the
observations were converted into FITS images using 10
millisecond time steps in order to obtain high time
resolution for this object. The primary objective was to
determine the brightness variation period.
Aperture photometry was used for each FITS image to
measure the “on” event rate. It was at the level of 0 most
of the time, however during blinks of the satellite, the rate
was increasing significantly. The event rate changes over the period of 2 minutes of observations are presented in
Fig. 22 and the flux rate changes of “on” events are
presented in Fig. 23
Figure 21. Event rate changes of "on" events from
Ajisai satellite during 2 minutes of observation on the
20th of November 2020.
Figure 22. Flux rate changes of "on" events from Ajisai
satellite during 3 seconds of observation on the 20th of
November 2020.
As seen in our data, the blinks are clearly visible and they
occur in a repeating pattern corresponding to the location
of mirrors on the surface of the rotating satellite body.
The data was analysed using Fourier periodogram with
the frequency of up to 2Hz.
Figure 23. Fourier periodogram up to frequency of 172800 c/d (2.0 Hz) for observations on the 20th of
November 2020.
The regularly distributed peaks visible in the Fig. 24 are
the harmonics of the main frequency. After further
inspection of event curves phased with the most
prominent frequencies we decided that the actual period
was P = 34885 ± 20.7 c/d as it presented the lowest scatter
of data points.
Similar analysis was carried out with subsequent
observations. Its result is presented in Tab. 3. We also
compared our determinations with long term ephemeris
published in [3]. Fig. 25 shows a perfect agreement
between expectation and our determination proving that
DVS camera can be a valuable tool in period
determination of fast rotating satellites.
UTC date observed
spin period
[s]
uncertainty
[s]
years
after
launch
ephemeris spin
period [s]
2020-11-20 2.4788s 0.0014s 34.275 2.4785
2021-02-13 2.4878s 0.0019s 34.508 2.4871
2021-02-15 2.4870s 0.0021s 34.513 2.4872
Table 3. Ajisai rotation period determined.
Figure 24. Three periods of the Ajisai satellite
determined during this project showing a clear sign of
growth, as expected from the ephemeris (dashed line)
from [2].
7 CONCLUSIONS
The possibilities of present day DVS cameras, as
concluded from testing the DAVIS 346BSI, in terms of ground-based observations of Earth satellites are
currently limited to some extent in several areas. As
opposed to modern CCD and CMOS cameras, the sizes
of available detectors are very small. Currently there are
not many specimens of the camera available off-the-
shelf, so it is not easy to find a camera with a matching
pixel size for a particular optical system. Moreover, due
to the fact that the DVS cameras are not cooled, the low-
light sensitivity is limited.
As a result of performing various tests on DAVIS 346BSI
camera, certain prerequisites have emerged, the fulfilment of which is necessary for performing survey
and tracking observations with the DVS. Firstly, the DVS
camera timing has to be related to UTC time scale. It is
possible through feeding the camera every few seconds
with an electronic rectangular impulse generated with
GNSS based clock. Thanks to this solution, the
conversion of internal timestamps to UTC is possible.
The second prerequisite is the conversion of the data
format generated by DVS cameras (AEDAT) into FITS
file format, so that it can be analysed with regular
astronomical photometric or astrometric software. For
this purpose, the AEDAT Converter software was
developed within the project.
Once the aforementioned requirements are fulfilled, it
was possible to perform several tests which shown that
the limiting magnitude of the DVS proved to be worse by
about 1 mag with respect to modern sCMOS sensors used
at the same telescope at angular velocity > 0.01°/s. At
velocities smaller than 0.01 deg/sec the difference
increases to about 3 mag. Based on the tests, the time bias
of the DAVIS 346BSI camera has the value at least
several milliseconds for bright targets and several dozen
milliseconds for dim targets. Moreover, the results of
GNSS observations indicated that the time bias can be
even larger, reaching up to about 125 milliseconds.
Moreover each recorded moving target produces a
“comet tail” which makes the image analysis more
challenging.
It was also noted that only objects, which are brighter
than about 8 mag are visible during tracking
observations, while targets down to about 12.5 mag are
detectable at angular velocities between 0.001°/s and
0.01°/s. Specific software or hardware modifications can
mitigate this problem, as the rotating shutter and variable
telescope tracking methods provided satisfying results.
Another important finding is the fact that when the
DAVIS 346BSI camera is combined with a typical
astronomical telescope, astrometry is impossible in most
of the cases due to lack of reference stars, which is caused
mainly by the small sensor size and reduced limiting
magnitude. The daytime limiting magnitude of the
DAVIS 346BSI camera is about 2-3 mag, which is worse
than in the case of most of the modern CMOS cameras.
This makes the daytime survey or tracking observations
with the DAVIS 346BSI camera (with the exception of
several brightest satellites) practically impossible.
The satellite tracking is additionally straitened by the
dependence of the DAVIS 346BSI camera time bias on
the target brightness. In the case of this type of
observations, the reference stars move in the field of view
with substantial angular velocities and each of them has
a different delay in DVS recording depending on its
magnitude. This may cause a deformation in relative
stellar positions that would have to be corrected.
Therefore, when a fast-moving object is observed, an
additional step in astrometric analysis is necessary for
DVS sensors when compared to regular CCD or CMOS
cameras. However, this effect is probably significant only
in the case of tracking the fastest MEO and LEO targets.
What is worth emphasizing is the fact that the limiting
magnitude of the DAVIS 346BSI camera is limited by
only 1-1.5 mag with respect to modern CMOS cameras
in satellite survey observations. It is possible that with the
future low-noise cameras, the on-the-fly detection and
analysis of objects passing in the field of view will be
achievable thanks to the DVS camera low bandwidth.
The time bias in astrometric measurements of GNSS
satellite performed with the DVS camera, was at the level
of ~125ms which is higher than ~4ms reported in literature [2] and ~20ms in our blinking LED
experiments.
The overall conclusion is that the current generation of
DVS cameras is a demanding tool, still at an early stage
of development considering its potential astronomical
use. The cameras were not developed to work under a
very low light condition. Perhaps with the exception of
Ajisai satellite rotation period determination,
observations with DAVIS 346BSI camera had no
advantage over a CMOS camera. In order to improve
DVS cameras’ results and poor astrometric success rate
either a much larger sensor size or much faster optical
systems would be necessary. Several improvements were
deemed beneficial for the DVS to be used in the future in its most promising area — satellite survey observations.
Those include using fast optical systems; implementing a
cooling module, thus significantly reducing the noise
events generation; lowering the detection threshold from
10% to at least 3%, introducing improvements to the
event stream conversion to FITS images or direct
analysis of AEDAT stream and finally reducing the pixel
size to match the camera properly with fast optical
systems.
Regardless of the rather early stage of technological
readiness of the DAVIS 346BSI camera, it is clear to see
the applications where this technology is worth consideration. DVS technology might be useful in
adaptive optics as a wavefront sensors camera. Another
interesting prospect for the use of the DVS technology in
the future is recording the lunar flashes, which are the
rapid changes in the brightness of the Moon’s surface
caused by meteorite strikes.
Considering the presented method of DVS data stream
conversion to 2D images (Fig. 26) it is obvious that we
do not utilize its full survey potential. In the case of
survey observations it would be beneficial to combine
events into images in the direction parallel to the target’s motion vector (Fig. 27). This method is equivalent to a
shift-and-add (synthetic tracking) method used, among
others, in very fast NEO observations. However, in
contrast to the shift-and-add, with the DVS cameras it is
possible to work with individual pixels and better
separate them from objects and background noise.
Obviously this method would have to be complemented
with a searching algorithm that would identify the
passage and calculate its direction.
Figure 25. Standard method of converting the DAVIS event stream to a 2D image. Events are selected
between two chosen moments of time and are
subsequently stacked onto and XY plane parallel to the
time axis.
Figure 26. Adaptive event stacking in which events are
combined into images along the target trail. A much
higher signal to noise ratio is possible since all events corresponding to a single object are grouped together in
just a few pixels.
8 REFERENCES
1. Cohen, G., Afshar, S., Schaik, A. et al (2017)
“Event-based sensing for space situational
awareness”, Proceedings Of The 18Th
Advanced Maui Optical And Space Surveillance
Technologies Conference, 19-22 September
2017.
2. Delbruck, Hu, He (2020) “V2E: From video
frames to realistic DVS event camera streams”,
arXiv e-prints.
3. Kudak V. (2017) “Studying of the own rotation
period changes of satellite "Ajisai" on the
interval 1986-2017”, Scientific Herald of
Uzhhorod University Series Physics.
4. Żołnowski M., Delbruck T., Kamiński K. et al. (2019) “Observational evaluation of event
cameras performance in optical space
surveillance”, ESA NEO SST Conference.
